Maximilian Schmitz-Foriest

Despite significant advances in robotics, most systems remain confined to specific tasks, environments, and embodiments, requiring extensive retraining when deployed in novel scenarios. This limitation severely restricts the scalability and economic viability of robotic solutions across diverse applications. This research addresses the fundamental challenge of knowledge transfer in robotics through a comprehensive investigation of transfer learning across three critical dimensions: robot embodiments, task domains, and environmental contexts.

Building on current progress in transfer learning, this will develop novel methods for cross-dimensional transfer that can handle unique robotic challenges including the sim-to-real gap, and negative transfer risks, and identifying appropriate abstraction levels for knowledge transfer. This research will contribute new algorithms for safe and efficient knowledge transfer, validated through experiments across multiple robot platforms and task domains. This work aims to advance the field toward truly general-purpose robotic systems capable of rapid adaptation to new scenarios.